Research map / Learning theory and algorithms / Clustering
K-Medoids For K-Means Seeding
James Newling, François Fleuret
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use CLARANS instead of k-means++ for initialization | Lloyd k-means · 23 datasets | initialization MSE: 30 % | Improved |
| Use CLARANS instead of k-means++ for initialization | Lloyd k-means · 23 datasets | final minimum MSE: 3 % | Improved |
| Use CLARANS swaps rather than medlloyd Voronoi iteration | K-medoids · synthetic and real datasets | energy | Improved |
| Add CLARANS after k-means++ initialization | Lloyd k-means · complete RNA dataset | mean final MSE: 0.83 -> 0.76 | Improved |
| Add CLARANS before Lloyd without k-means++ pre-initialization | Lloyd k-means · complete RNA dataset | mean final MSE: 0.83 -> 0.76 | Improved |
| Replace k-means++ initialization with MCMC initialization | Lloyd k-means · 23 datasets | initialization MSE | No clear effect |
| Replace uniform initialization with k-means++ | Lloyd k-means · 23 datasets | initialization and final MSE | Improved |
| Replace uniform initialization with the bf scheme | Lloyd k-means · 23 datasets | final MSE | No clear effect |
| Use optimization level 2 rather than level -2 (ablation) | CLARANS · simulation data, σ = 2⁻⁴ | log₂(distance calculations): 44.1 -> 26.7 | Improved |
| Use optimization level 2 rather than level 0 (ablation) | CLARANS · simulation data, σ = 2⁻⁴ | time [s]: 407 -> 15.6 | Improved |
| Add level 3 subsampling to exact CLARANS accelerations (ablation) | CLARANS · N = 500,000, d = 4, K = 500 Gaussian data | convergence time: within 20 seconds | Improved |
| Pre-initialize CLARANS with k-means++ (ablation) | CLARANS + Lloyd · complete RNA dataset | mean final MSE: 0.76 -> 0.76 | No clear effect |
Closest papers by meaning
- k-variates++: more pluses in the k-means++ (ICML 2016 · similarity 0.88)
- Fast and Accurate $k$-means++ via Rejection Sampling (NeurIPS 2020 · similarity 0.87)
- Approximate K-Means++ in Sublinear Time (AAAI 2016 · similarity 0.87)
- Analyzing $D^\alpha$ seeding for $k$-means (ICML 2024 · similarity 0.86)
- OneBatchPAM: A Fast and Frugal K-Medoids Algorithm (AAAI 2025 · similarity 0.84)
- Near-Linear Time Approximation Algorithms for k-means with Outliers (ICML 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
